Low-gain Control Strategy for Robot Manipulators Based on Sparse Feature Learning Dynamics with an Application to Collision Detection

Recently, with the robotic technique developing toward precision and intelligence, robots have been used more widely in human life and production. As a common feature in the foreseen applications, robots should be able to detect unexpected collisions while ensuring dynamic accuracy, so as to improve safety at work. In the previous model-based collision detection solution, most methods assume that the robot dynamic model is complete and accurate. Unfortunately, reliable robot dynamics is hard to obtain due to model uncertainties, assembly errors, and the lack of information provided by manufacturers. This paper proposed a novel low-gain control strategy based on sparse feature learning dynamics. Firstly, without considering the physical structure parameters, the dynamics was learned directly via the data-driven technique. Secondly, according to the learned accurate dynamics, a model-based low-gain controller was designed to ensure control performance while avoiding excessive unspecified force. Finally, using this control strategy, sensorless collision detection was realized in a 7-DOF manipulator and the performance of the proposed method was evaluated.